This is not the AI we were promised | The Royal Society

The Royal Society Michael Faraday Prize Lecture for 2025 is delivered by Professor Michael John Wooldridge, focusing on the current state and limitations of contemporary artificial intelligence systems.

The lecture addresses the impressive capabilities of modern AI, such as large language models like ChatGPT, contrasting their perceived expertise with their fundamental failures in demonstrating rational intelligence. Professor Wooldridge discusses specific shortcomings, including an inability to discern truth from falsehood, inherent inconsistency, a lack of self-awareness regarding their knowledge boundaries, high suggestibility, and a tendency toward fantastical outputs.

These phenomena are compelling AI researchers to establish new experimental fields to understand and manage these complex artificial artifacts, prompting a re-evaluation of the initial promise of truly intelligent machines.